Deterministic Strategy

A deterministic strategy based upon the decision network involved the Bayesian Poker Player performing the action which had the highest EW at that point in the game. The results in Figure 7 show that in testing against the earlier version of BPP which used betting curves to select an action, the deterministic version performed extremely well, consistently beating its opponent and accumulating winnings at a rate of $0.2922 \pm 0.0236$ betting units per game ( $t = 12.3734; p \le 0.025$). These gains may have been in part due to the lack of effective opponent modeling evident within the earlier model, which was unable to identify and exploit the predictable deterministic strategy employed. As a result of this weakness, a more ``optimal'' player would clearly have the advantage over a randomised model and the results of the self-play experiments should not be considered too heavily when assessing any improvement in performance. A better assessment of the performance merits of a deterministic strategy would be gained through competition with experienced human opponents.

  

Figure 7: Comparitive performances of BPP implementing a decision network.